fitrialif / nsga-keras

evolution for NAS based on NSGA/NSGAII/NSGAIII (with parallel evaluation)

Neural Architecture Search for Keras Sequential Models

Authors: Petra Vidnerová, Štěpán Procházka. The Czech Academy of Sciences, Institute of Computer Science.

Genetic NAS based on NSGA, NSGAII or NSGAIII algorithms. Works for Sequential models only.

Requirements:

Tensorflow, Keras, Deap, numpy, pandas, Scikit-learn

Usage:

TODO: move complete setup to one configuration file

Example:

Docker

To run project code independently of your own system setup, Dockerfile with convenient docker-compose project is provided. Inside the container all the required stuff is installed and using the docker-compose , project code is mounted for convenient exchange of data between container and the underlying machine.

Make sure you have docker and preferably docker-compose installed. Assuming that current working directory is the root of the repository, use following commands to run the project.

# runs docker container with jupyter lab server (available on `localhost:8888` by default)
docker-compose up

# terminate the container and free its resources
docker-compose down

# run command (e.g. interactive shell) inside the container
docker-compose exec nsga-keras /bin/bash

# builds the image (or rebuilds it if previous version exists)
docker-compose build

TODO: add support for GPU inside docker